Kazbrekker
A sentiment classifier fine-tuned from BERT-base, sorting text into Positive, Negative, Neutral and Irrelevant. Published on Hugging Face, with 11 downloads in its first 12 hours.
- accuracy
- 92.2%
- macro F1
- 0.922
- val loss
- 0.334
Manas Ranjan Bissoyi · Founder, Ateegnas
I'm building Ateegnas, an AI-native annotation layer that delivers consistent, high-quality training data at speed.
Teams spend months tuning models and comparatively little on what goes into them. Mislabelled samples, thin coverage and noisy collection show up later as a product that fails in ways nobody can trace.
My belief is simple: the weakest link in any AI product isn't the model, it's the data feeding it. That's the problem I'm solving.
An AI-native annotation layer, built from Nabarangpur, Odisha. Ateegnas produces consistent, high-quality training data at ultrafast speed, so AI teams stop losing accuracy to noisy labels.
I taught myself AI workflows by building: RAG-based tools, a fine-tuned BERT model on Hugging Face, and full-stack products.
A sentiment classifier fine-tuned from BERT-base, sorting text into Positive, Negative, Neutral and Irrelevant. Published on Hugging Face, with 11 downloads in its first 12 hours.
A leaderboard where companies bid with startup credits, discounts and goodies to stay on top, so founders find the best perks in one place.
A persistent memory layer for AI agents, so they keep context across sessions. Includes a developer dashboard for what an agent stores and recalls.
Affiliate marketing, crowdfunding and student counseling came first. Then three startups and a 300-founder community before Ateegnas.
Collected egocentric data in the scholars program, a first-hand look at how large annotation teams gather real-world data.
Building data infrastructure for AI.
A short-drama platform for affordable, episodic entertainment, designed to pay creators more and keep viewers engaged.
Started and grew “Fund Your Idea” to 300+ startup founders in under three months and ₹60K in revenue, sharing grant and funding databases, tools and pitch-deck support.
An organic social platform for genuine local connection, with no fake accounts or theme pages.
Guided students from many boards and regions through course selection and onboarding.
Ran social-media crowdfunding campaigns and made the content that brought in donors. Interned at Muskurahat Foundation over the same summer.
Sold courses on Canva, Facebook Ads and Google Ads through affiliate campaigns with a high conversion rate.